Article: Tang, C., S. Hamilton, and E.F. Chang. “Intonational Speech Prosody Encoding in the Human Auditory Cortex.” Science 357, no. 6353 (2017): 797–801. DOI: 10.1126/science.aam8577

Assignment: Read the assigned article carefully (expect this to take at least two hours, possibly more) and think about it, then provide short answers to the questions below. You can do this by responding straight on this document or by submitting a separate document with numbered answers. Be clear and concise, points will be taken off for unnecessary words.

A few things that are not well explained in the paper:

• I didn’t see a definition of H gamma (indicated as the Greek letter which I cannot get my text editor to make) in the figures – this is just “high gamma power”, which is the main measure of neural response used in this paper and in other intracranial recording studies.
• When the paper refers to “spectral information” (which I highlighted in orange), that just refers to different amounts of power at different frequencies. See the “spectrogram” in Figure 1A, which shows auditory power (darkness) as a function of time (x axis) and auditory frequency (y axis).
• You can ignore the stuff about the “missing fundamental” (highlighted in orange).

TA tip: when reading, try to focus on the figures and make sure you understand them. Refer back to the text for supporting information when needed.

Intonational speech is emphasizing particular words. This can change the meaning of a sentence.

1. What method does this paper use?

high-density electrocorticography

1. What is the main (most important) empirical finding in the paper? What figure illustrates this finding?

Neural populations measured by electrodes selectively represented intonation contours normalized by speaker (through using relative pitch instead of absolute pitch). Figure 4 shows this. The female and male voice differ significantly in absolute pitch but not relative pitch. The neural representations are not distinguishable.

1. What is the design of this experiment? That is, what factors were manipulated (list each), and how many different conditions were used?

The main independent variable is the intonation contour. Three different types of linguistically significant contours was used. Speaker identify and absolute pitch were varied, but duration and intensity were kept constant.

1. What does Figure 1E show?

It shows that male and female voices show identical cortical activity despite the significant difference in absolute pitch.

1. The middle graph in figure 2N shows 147 electrodes that had information about which sentence was spoken. Could you decode from these electrodes whether the speaker saying those sentences was male or female?

No because all other variables have a low proportion of R2 for these electrodes.

1. Would you expect the electrodes that can discriminate intonation to be able to discriminate musical melodies (in which each note has a different pitch)? If so, how would you characterize the function of those sites?

An interesting experiment is to play the song song in different keys. This is analogous to different speakers absolute pitches. Melodies are based on pitch contour not absolute pitch so we can expect these electrodes to discriminate. However it’s possible that another mechanism makes these neurons selective to speech only.